Labrador Sea Water (LSW) is a principal convectively-formed water mass of the subpolar North Atlantic (SPNA). Using extensive oceanographic archives we demonstrate striking changes in SPNA caused by massive LSW production during 1987–1994 and document recent salinification and warming, imminently bringing SPNA to the state last time seen in the 1960s. Two prominent LSW classes are spreading across SPNA since the 1980s. The first, record dense, deep, and voluminous, class has been progressively built by intense winter convection through 1987–1994. Even though most of this LSW has left SPNA, its remnants are still present there. The second, shallower, class strengthens in 2000; over subsequent years its core becomes slightly thicker and deepens. The anomalous signals acquired by these LSW classes in their formation region arrive in the Irminger and Iceland basins with the characteristic delays of two and five years for deeper LSW and a year and four years for shallower LSW.
 By governing the meridional overturning circulation (MOC) of the Atlantic Ocean [Dickson et al., 2002; Bryden et al., 2005], the northern North Atlantic (NA) plays an active role in the ventilation of the deep layers of the whole world ocean. Its most remarkable intermediate water mass is known as the Labrador Sea Water (LSW). LSW is formed in the Labrador Sea through deep convection caused by high heat losses during severe winters [Lazier, 1980; Clarke and Gascard, 1983; Gascard and Clarke, 1983]. This process, which is believed to be controlled by the phase and persistence of the North Atlantic Oscillation (NAO) [Dickson et al., 1996], turns the Labrador Sea into a main contributor of newly formed or modified intermediate waters to the Atlantic MOC. The NAO index is the normalized Azores-to-Iceland sea level pressure difference, linked to the strength of the large-scale zonal atmospheric transport [Hurrell et al., 2001].
 When produced in large quantities, LSW spreads across the subpolar North Atlantic (SPNA) [Talley and McCartney, 1982], filling its intermediate reservoir with water that is relatively fresh, cold, and rich in dissolved gases [Sy et al., 1997; Yashayaev and Clarke, 2006; Kieke et al., 2006]. LSW regulates the large-scale dynamics of NA in three ways. First, its volume and property changes influence the mid-depth circulation, mixing and signal propagation in the subpolar basins. Second, varying LSW production controls exchanges between the subtropical and subpolar gyres at their intermediate depths [Curry and McCartney, 2001]. Third, by contributing to the intermediate compartments of NA, LSW affects the properties [Dickson et al., 2002] and controls the vigor (K. Boessenkool et al., On the relationship between the North Atlantic Oscillation and deep-ocean flow speed changes during the last 230 years, submitted to Geophysical Research Letters, 2007) of the lower limb of the Atlantic MOC, originating from the cold and dense polar overflows entering NA via the deep trenches in the Greenland-Scotland Ridge. The Iceland-Scotland Overflow Water (ISOW), evolving into Northeast Atlantic Deep Water (NEADW), and Denmark Strait Overflow Water are the two major deep dense overflows arriving to NA from the Nordic Seas. They pass through and mix with LSW just before they enter the deep and abyssal NA reservoirs, entraining fresh water, elevated gas, and transient tracer signatures of LSW, ultimately reaching the deeper layers of the Labrador Sea where their evolving properties are being monitored [Yashayaev et al., 2003].
 The convective cooling and freshening of the Labrador Sea over the 1980s and early 1990s have produced LSW that by 1994 became the coldest, densest, deepest and most voluminous since the 1930s [Yashayaev et al., 2003]. The LSW pathways shown in Figure 1a are based on the thicknesses of the LSW layer measured over three years following the most massive production of this water mass (dashed lines). In the present article we discuss the changes in the subpolar gyre triggered by extremely deep convective mixing in the early 1990s that led to production of this exceptional LSW and document some footprints of its development and transformation.
Pickart et al.  argued that LSW is likely formed in the southwest Irminger Sea, pointing at the fact that tracer observations by Sy et al.  yielded very high advective speed and short travel times for LSW implying that it takes about six months for the water to spread to the Irminger Sea, while an advective-diffusive model study by Straneo et al.  showed that LSW arrives in the central Irminger basin two years after its formation. To explain such a discrepancy in views Pickart et al.  rejected the assumption that the sole source of LSW is the Labrador Sea and “formed” LSW in the Irminger Sea thus explaining the seeming too-fast spreading of this water mass. However, the LSW transit times based on a short observational record by Sy et al.  has not yet been questioned. Here we provide observational evidence that the LSW transit times are notably longer than those suggested before, justifying the advective-diffusive nature of LSW spreading and reinstating the dominant source of LSW in the Labrador Sea.
Figures 1a and 2(inserts) indicate hydrographic stations in close proximity of repeat trans-Atlantic section AR7. First occupied under the aegis of the World Ocean Circulation Experiment (WOCE, 1990-1997), this section later became the principal observational element in the Atlantic Climate Variability and Predictability (CLIVAR). Its Labrador Sea part, AR7W, has been occupied by the Bedford Institute of Oceanography (Canada) since 1990, while the eastern part, AR7E, has been surveyed by the University of Hamburg, the German Hydrographic Office, and Royal Netherlands Institute for Sea Research since 1991. To portray LSW in a year typical for the low NAO phase and highlight the outstanding contribution of the 1990s to the multi-decadal history of LSW production and rapid restratification of the mid depths during the last decade, we compare AR7 occupations with a composite section built as 1966 replica of AR7, which mainly includes the most extensive SPNA survey before WOCE led by John Lazier (CSS Hudson, 1966).
3. Identification of LSW
 All distinguishable instances of LSW form distinct peaks in volumetric density (σ2—potential density anomaly referenced to 2000 dbar) and temperature-salinity (T-S) censuses of each basin-survey. The σ2-T-S range, height, and shape of such a peak identify a unique LSW, while the peak's maximum points to the core of this water mass. The AR7 T-S projections will be published separately; here we introduce a σ2layer volumetric approach. More compact than its T-S companion, the σ2 method is equally effective for the LSW identification-both methods produced identical σ2 ranges for the LSW cores discussed in the paper. The essence of the σ2 volumetric approach can be expressed by plotting basin-mean thickness of individual σ2 layers (Δσ2 = 0.01 kg m−3) in the σ2-time coordinates. Such a diagram for the Labrador Sea is presented in Figure 1b. It was constructed by averaging layer thicknesses from individual hydrographic stations (in a single year or survey) weighted by the distance or area represented by these stations. At each station of a basin-survey each LSW was defined by a σ2 range (±0.017 kg m−3) centered at the volumetric core (Figure 1b; Labrador Sea, e.g., 36.916∣1990, 36.940∣1993). These individual density ranges collectively form an LSW historic density range. All volumetric peaks in Figure 1b can be grouped into two continuous progressions, each reflecting year-to-year development and transformation of a given LSW core. This grouping introduces an LSW class–a sequence of LSW types linked by the same development history.
 The 1980-2006 AR7 repeats allowed us to build the complete history of development, evolution and decay of two prominent LSW classes in SPNA. The first class, LSW1987–94 (the subscript indicates a time interval over which this LSW had likely been formed), is associated with the most extraordinary documented LSW production (Figure 1b). Record voluminous in 1994, it has strongly diminished over the past twelve years, recently becoming barely identifiable in the volumetric diagrams. The other class, LSW2000, was formed in 2000. It is shallower and has lower densities than LSW1987–94. We name these two classes after the years when they achieved their coldest, densest (note temporal σ2 changes in each LSW) and most voluminous states.
 When a certain LSW class loses its volumetric prominence (recent LSW1987–94) other criteria can be used to refine its volumetrically defined σ2 range. Among these are salinity (Figure 2), temperature [Lazier et al., 2002] and potential vorticity [Talley and McCartney, 1982] minima, oxygen maximum [Clarke and Coote, 1988] and the anthropogenic tracers [Azetsu-Scott et al., 2003]. Even when thinning remnants of highly modified LSW1987–94 formed weak volumetric maxima, the mentioned extrema remained within the volumetrically defined σ2 ranges, thus confirming or endorsing the volumetric identification of strongly transformed LSW.
 We disagree with the approach taken by Kieke et al. , introducing “classical” and “upper” LSW that are defined by potential density ranges of 27.74–27.80 and 27.68–27.74 kg m−3, respectively. If combined, these two layers fill most of the top 2000–2200 m of the Labrador Sea leaving no room for other intermediate waters arriving form outside of the Labrador Sea to replace LSW. We argue that LSW does change its density (Figure 1b), occasionally “crossing” the 27.74 kg m−3 boundary. On the other hand, the density range defining classical LSW is excessively broad, indicating that this water was never thinner than 1000 m (at odds with our LSW1987–94 findings), and thus systematically overestimating the LSW thicknesses and production rates. We also find it inappropriate to label the densest and deepest LSW unique for the 1990s as “classical”.
 The AR7 section plots (e.g., Figure 2) and volumetric inventories (e.g., Figure 1b) demonstrate that a universal definition of a given LSW class (e.g., LSW1987–94) meant to be used through its development and transformation history should not be based on a static range of a certain variable. Intensifying and deepening convective mixing, followed by water mass dislodging and diluting do change the properties of LSW and, as a result, alter its core and boundaries. This means that any criteria involving the LSW properties must account for possible changes in such properties. The volumetric method used in our study automatically adjusts to a specific LSW core, following its year-to-year dynamics and transformation and revealing spatiotemporal changes in this LSW.
4. Salinity on the Trans-Atlantic Section North of 50°N
 The compilation of composite trans-Atlantic hydrographic sections for 1966, 1994, and 2004 (Figure 2) introduces all principal water masses of SPNA, reflecting transitions from their saltiest (and warmest) state of the mid-late 1960s to extremely fresh (and cold) phase of 1994 and then to generally saltier (and warmer) conditions of the recent years. A distinct intermediate salinity minimum associated with LSW can be recognized in all sections crossing the subpolar basins. However, the depth, thickness, and properties of this water change from basin to basin, within basins and between surveys.
 The year 1966 (Figure 2a) is characteristic for record salty and warm SPNA of 1964–1971 (since the 1930s) [Yashayaev et al., 2003; Yashayaev and Clarke, 2006]. This year is also central in the 1960–1971 low NAO phase. The winter of 1965–1966 was particularly mild in the Labrador Sea [Lazier, 1980]. Thus it was unlikely that significant convective renewal of LSW occurred during the mid-late 1960s, causing the LSW lying at the intermediate depths to remain isolated from the upper layer and become saltier and warmer through its mixing with surrounding waters. In 1966 deep LSW could be identified in the Labrador Sea as a nearly homogeneous layer with salinities between 34.88 and 34.90 (in the distance range −700 − −240 km). A retrospective analysis suggests that the last significant LSW renewal of the 1960s occurred in the winter of 1962–1963 [Lazier, 1980].
 A particularly large change occurred between the 1966 and 1994 surveys. Indeed, in 1966 the deep LSW core was everywhere saltier, warmer and shallower than in any hydrographic survey of SPNA during the 1990s. This change is a result of production of an exceptionally cold, fresh, dense, deep, and vertically homogeneous LSW class by strong winter convection between the late 1980s and the early-mid 1990s [Lazier et al., 2002]. This water was, in fact, the most voluminous LSW in its historic 70-year record–LSW1987–94. In 1994 this water mass appears as the most prominent feature of the intermediate layers, filling the entire central part of the Labrador Sea basin within the depth range of 500−2400 m (Figures 1b and 2b). This means that within the Labrador and Irminger basins and to some degree in the Iceland Basin the well-mixed body of the fresh LSW1987–94 has penetrated to the depths earlier occupied by NEADW. As a result, this LSW exceeded both vertically and horizontally any other water mass seen in SPNA since the 1930s.
 As time progressed, temperature, salinity and density stratifications re-established above the thinning patch of LSW1987–94. Such isolation of LSW1987–94 was a result of a substantial decrease in the net annual heat loss from the Labrador Sea to the atmosphere after 1994 [Lazier et al., 2002; Yashayaev and Clarke, 2006]. By 2004 (Figure 2c), most of the excess volume of LSW1987–94 has disappeared and the water columns have restratified above its deep core causing stratification in the whole SPNA to change. The saltier and warmer remnants of LSW1987–94 could still be recognized in 2006 (not shown). In the Labrador Sea, its present signature is a weak local salinity minimum between 34.88 and 34.90, and the slightly increased spreading of the isotherms between 2.9 and 3.1°C (not shown). By 2004 the patch of LSW1987–94 in the Iceland Basin had also started to lose its volume and gain salt.
 In 2004, the Labrador and Irminger Seas also exhibit two new (to 1994) features. The first feature is a homogenous layer between 400 and 1300 m. This water corresponds to the volumetric density class LSW2000 (Figure 1b). It was massively formed by winter convection in 2000 and has been then modified by mixing and moderate convection during subsequent years [Yashayaev and Clarke, 2006]. Even though some shallow mixed layers could be found in the Labrador and Irminger Seas since 1997, it was only the winter of 1999–2000 when this water developed into a distinct and homogeneous LSW class, since then maintaining its integrity. The increase in winter convection in 2000 coincides with five-year high NAO, pointing at high heat losses in the Labrador Sea and explaining why the convectively formed water spread deeper and wider in 2000 than in the 5 previous years.
 The second new feature of the 2004 section (Figure 2c) is the relatively salty and warm intermediate layer separating LSW1987–94 and LSW2000 within the Labrador and Irminger Seas. It is the core of saltier and warmer water arriving to replace the deeper LSW. This water originates from Icelandic Slope Water (ISW) seen near the Reykjanes Ridge. ISW in its turn is formed through a direct linear mixture of the original Iceland-Scotland Overflow Water with the overlying Atlantic thermocline water near the Faroes, without interfering LSW [van Aken and de Boer, 1995]. That ISW then follows the slopes of Iceland and the Reykjanes Ridge until it enters the Irminger Sea through the Charlie-Gibbs Fracture zone. From the western slope of the Reykjanes Ridge the ISW intrudes into the centre of the Irminger gyre, forming a relatively thin, but salty and warm layer, now prominently seen between the LSW1987–94 and LSW2000 cores. This characteristic salinity maximum is typically 140–200 m deeper than its temperature companion.
 In 2004 a clear LSW2000 salinity minimum was first observed in the Iceland Basin.
 Summarizing the long-term changes in the LSW properties based on hydrographic sections backed by time series [Dickson et al., 2002; Yashayaev et al., 2003; Yashayaev and Clarke, 2006], we report a strong salinity and temperature contrast between the 1966 and 1994 sections (these two extreme years set boundaries for the historic ranges of the LSW properties). We also state that the intermediate layers of SPNA are now approaching the salinity and temperature levels of 1966 (typical for record salty and warm state of the mid 1960s – early 1970s). This is consistent with the recent decline in NAO and winter convection.
5. Transit of LSW Anomalies to the Irminger Sea and Iceland Basin
Figure 3 shows potential temperature (Figure 3, left) and salinity (Figure 3, right) anomalies in the two LSW classes — LSW1987–94 (Figure 3, bottom) and LSW2000 (Figure 3, top) and adjacent to them warmer and saltier waters. The distance is measured along the composite AR7 section and referenced to the center of the Labrador Sea; fractional year reflects actual date of each station.
 As already stated in our discussion of Figure 1b, the development of deep convection in the Labrador Sea between 1987 and 1994 resulted in consecutive annual increases in the LSW density during these years. A series of papers describes the mechanisms of such buoyancy loss and documents the temperature and salinity changes accompanying the intensification of winter convection in the Labrador Sea [Lazier et al., 2002; Yashayaev et al., 2003].
 Between 1987 and 1994 LSW1987–94 was rapidly changing within its source region in the Labrador Sea, cooling by about 0.45°C (Figure 3, left bottom), becoming 0.06 kg m−3 denser and doubling in volume (Figure 1b). Note that during all these years except the last, 1994, LSW1987–94 was steadily becoming saltier (Figure 3, right bottom, Labrador Sea, 1988–1993). This change in the LSW1987–94 salinity means a loss in the LSW buoyancy in addition to that caused by the multiyear cooling of LSW1987–94 (Figure 3, left bottom, Labrador Sea, 1987–1994). To remind, this LSW cooling was a result of extreme heat losses to the atmosphere during the severe winters of the early 1990s [Lazier et al., 2002]. These matching tendencies in the temperature and salinity effects on the LSW density explain the rapid increase in the LSW density (Figure 1b). The noted increase in the LSW1987–94 salinities was due to mixing with the saltier NEADW that is entrained into LSW every time when convection gets deeper. The progressive deepening of the lower margin of LSW and therefore of winter convection was observed between 1988 and 1993 [Yashayaev and Clarke, 2006], supporting our explanation of the salinity increase. The last occurrence of intense convection 1993/1994 [Lazier et al., 2002] did not seem to reach deeper than in 1993, simply because in the spring of 1994 the depth of the mixed layer was not noticeably deeper than in the spring of 1993 [Yashayaev and Clarke, 2006]. Therefore, it is most likely that winter convection of 1994 did not bring up enough of salty NEADW from below LSW to overcompensate the freshening effect of convective entrainment of the upper low–saline waters. This has resulted in a single year disruption of the LSW salinity increase. Until 1993 this salinity increase was maintained by the NEADW entrainment and after the cessation of LSW1987–94 convective renewal in 1994 by mixing with other intermediate waters (saltier than LSW). That 1993 to 1994 salinity decrease by almost 0.01 has formed the second minimum in the bi-minimum LSW1987–94 salinity record. However, newly formed LSW was colder in 1994 than it was in 1993, thus not disrupting or even slowing the tendency of its cooling heading to the all-time temperature low seen in 1994. Since 1994 the deep reservoir of LSW1987–94 has mainly remained isolated from the winter mixed layer [Lazier et al., 2002], steadily becoming warmer and saltier and showing only slight density changes (Figures 1a, 2b, 2c, and 3).
 The single temperature and dual salinity minima described above combined with other characteristic points in the LSW1987–94 development and transformation observed within the formation region of this water mass (in the Labrador Sea) serve as effective markers or tags, which arrival can be sought in the other Atlantic basins. For example, the buildup and following rapid decline of LSW1987–94 class expressed in the increase and then decrease in the corresponding density layer thickness (Figure 1b) have had their 2-year delayed imprint in an analogous volumetric compilation for the Irminger Sea.
 Linking the LSW1987–94 signals observed in the LSW formation region with the signals arriving in the Irminger Sea and Iceland Basin (Figure 3, compare with LSW pathways in Figure 1a) one can come to the following conclusions:
 1. The LSW1987–94 temperature anomalies were record low in the Labrador Sea in 1994, while the coldest LSW invaded the entire Irminger Basin only by 1996, implying a two-year delay. Since the bulk of LSW typically arrives to the Irminger Sea between 500 and 700 km, we primarily use this distance range to define the LSW transit time to the Irminger Sea.
 2. The sustained cooling of LSW1987–94 in the Iceland Basin ended only in 1999, five years after this water reached its coldest point in the Labrador Sea. Since 1999, this deep LSW of the Iceland Basin shows a slight warming.
 Not only extreme points were used to find the LSW transit times. We have also analyzed the shifts or transitions between extensive warm and cold LSW phases and documented the basin-to-basin transit of these phases. Here are some characteristic points of these warm-to-cold and cold-to-warm LSW1987–94 phase transitions: in the Labrador Sea cooling in 1990 and warming in 1998; in the Irminger Sea cooling in 1992 and warming in 2000; in the Iceland Basin cooling in 1995 and warming in 2005. The delays in the noted transitions confirm our estimates of transit times to the Irminger and Iceland basins (∼2 and ∼5 years, respectively).
 It was beyond our expectation to discover double salinity minima in the LSW1987–94 series for the Irminger and Iceland basins. However bi-minimum patterns similar to that seen in the Labrador Sea (but of weaker magnitude) can be recognized in the two other basins. Focusing only on the second low in the LSW1987–94 salinities, we consecutively see it in the Labrador, Irminger and Iceland basins in 1994, 1996, and 1999. This again confirms the two and five year transits of LSW1987–94 to the two latter basins.
 In addition to the spreading times, Figure 3 (bottom) also illustrates changes in the spatial extent of the LSW1987–94 bodies in the Labrador and Irminger Seas. LSW1987–94 expands spatially when convection intensifies and contracts when the deep intermediate layers lose LSW and restratify. Between 2000 and 2003 LSW1987–94 withdrew from the eastern part of the Irminger Sea leaving it to warmer and saltier waters, which include ISW, Subpolar Mode Water [McCartney and Talley, 1982] and also strongly modified LSW. For almost a decade now, the LSW1987–94 layers of the Irminger and Labrador Seas have been showing sustained warming and salinification. These signals can be tracked back to the western flank of the Reykjanes Ridge, where in 2000 a strong positive anomaly was first seen replacing a strong cold and fresh anomaly associated with LSW1987–94 (this transition is also reflected in the sections in Figure 2, west of Reykjanes Ridge).
 LSW2000 has been sporadically renewed during its short history, resulting in patchier temperature and salinity anomaly fields than those for the deeper water (Figure 3). Even if our sections and volumetric estimates suggest that LSW2000 and its anomalies arrive from the Labrador Sea, in some years this water could be remixed and its signals altered outside of the formation region. What assures us in this LSW spreading east is that since 2000, when it was first voluminously appeared in the Labrador Sea (Figure 1b), this water mass is steadily becoming warmer and saltier, showing its record low temperature and salinity in 2000. The corresponding developments of LSW2000 in the Irminger Sea and the Iceland Basin (Figure 3, top) suggest that it took, respectively, a year and four years for this water mass to reach each of these basins in order. These transit times are by about a year less than those for LSW1987–94. Note the westward spreading of the warmer and saltier anomaly in the LSW2000 layer from the flanks of the Reykjanes Ridge, to the Labrador Sea very similar to that in the LSW1987–94 layer. Finally, since the year of its first massive production (2000) the core of this water shows progressive deepening in both Labrador and Irminger Seas.
6. Discussion and Concluding Remarks
 Temperature and salinity anomalies measured in the cores of LSW1987–94 and LSW2000 (Figure 3) advect to the east reaching the Irminger Sea in two years and one year, respectively, and arriving in the Iceland Basin in five and four years after their formation in the Labrador Sea. It is not surprising that these transit times are at least double those suggested by Sy et al. , because the latter were based on a rather short observational record insufficient to register the true arrival of LSW into the two eastern basins. Our records do not show any evidence of the deeper LSW class (LSW1987–94) formed outside the Labrador Sea. This fact reinstates the true source of this water in the Labrador Sea. The volume of LSW is notably larger in the Labrador Sea than elsewhere, meaning that its fresh and cold anomalies, especially when associated with larger volumes, may result in comparable or even larger anomalies downstream. This explains why the observed LSW anomalies “preserve” their magnitude after leaving the Labrador Sea.
 In addition to showing new vintages of LSW spreading across the ocean, Figure 3 reveals the source and spreading of the recent warming and salinification of the mid depths. The westward pointing arrows in this figure (Figure 3) underline the essence of the three-dimensional exchange between the Labrador and Irminger basins maintained by the cyclonic circulation within each basin. After most of LSW1987–94 has drained from the Labrador Sea, the anomalously warm and salty waters entering the Labrador-Irminger gyre from the east and southeast (e.g., ISW arriving from the Reykjanes Ridge) become noticeable and their pathway can be mapped (Figure 3). This indicates that the Irminger-Labrador gyre receives waters from multiple sources and passes their anomalous features in both eastward (LSW) and westward (e.g., ISW) directions.
 The authors thank three anonymous reviewers for valuable comments and suggestions and express their gratitude to Allyn Clarke, John Lazier, Jens Meincke, Bob Dickson, and many others who over five decades surveyed, explored, and monitored the subpolar basins. The 2004 A1E hydrographic data are courtesy of Detlef Quadfasel and John Mortensen, and were collected with funding by the Bundesminister für Bildung und Wissenschaft (German CLIVAR) and the EU Commission (ASOF-E). Finally, many friends and colleagues of the authors contributed to this study by sharing their best skills and spirits on land and, especially, at sea.